# ParseBench: document parsing for AI agents (LlamaIndex leaderboard) (ParseBench)

> Scores document parsers and vision-language models on 2,078 human-verified enterprise pages across tables, charts, content faithfulness, semantic formatting and visual grounding, with the listed cost per page.

Canonical page: https://benchlm.ai/benchmarks/parsebench

- Category: [Multimodal & Grounded](/multimodal-grounded)
- Last updated: October 25, 2026 snapshot

## About ParseBench

- Year: 2026
- Tasks: 2,078 human-verified enterprise pages from insurance, finance and government documents
- Format: Deterministic test rules per page; five dimension scores and their mean, in percent
- Difficulty: No method is strong on all five dimensions; charts and visual grounding separate the field
- Paper: [ParseBench: A Document Parsing Benchmark for AI Agents](https://arxiv.org/abs/2604.08538)

ParseBench judges a parser on whether its output keeps the structure an agent needs rather than on text similarity. Each page carries deterministic test rules in five dimensions: tables (structure, merged cells, hierarchical headers), charts (data points, series, axes), content faithfulness (omissions and hallucinations), semantic formatting (strikethrough, superscripts, emphasis, heading levels) and visual grounding (element localization and classification). The overall score is the mean of the five dimension scores. The leaderboard lists every configuration LlamaIndex ran, from commercial parsing APIs and cloud document services to proprietary and open-weight vision-language models at each reasoning setting and local open-source parsers, with cost in cents per page where a list price exists. We mirror the leaderboard CSV from the ParseBench repository, which the public site loads at runtime, and store each catalog model's best configuration on its row.

ParseBench is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (149 configurations)

| Rank | Model | Configuration | Creator | Score |
|------|-------|---------------|---------|-------|
| 1 | [LlamaParse Agentic Plus](https://www.parsebench.ai/#leaderboard) | LlamaParse · Default · 5.63¢ per page | LlamaIndex | 88.82% |
| 2 | [LlamaParse Agentic](https://www.parsebench.ai/#leaderboard) | LlamaParse · Default · 1.25¢ per page | LlamaIndex | 85.90% |
| 3 | [Pulse Ultra 2](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Default · 1.50¢ per page | Pulse | 81.60% |
| 4 | [Upstage Document Parse 2 (Enhanced)](https://www.parsebench.ai/#leaderboard) | Proprietary VLM · Enhanced · 3.00¢ per page | Upstage | 80.51% |
| 5 | [anyformat (Standard)](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Standard · 4.30¢ per page | anyformat | 80.30% |
| 6 | [LlamaParse Cost Effective](https://www.parsebench.ai/#leaderboard) | LlamaParse · Default · 0.38¢ per page | LlamaIndex | 79.62% |
| 7 | [Claude Opus 5.5](/models/claude-opus-5-5) | Proprietary VLM · Effort High · 6.13¢ per page | Anthropic | 78.89% |
| 8 | [oi-parser](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Default · no listed cost | — | 78.30% |
| 9 | [Claude Fable 5.1](/models/claude-fable-5-1) | Proprietary VLM · Default · 16.05¢ per page | Anthropic | 78.12% |
| 10 | [rakedoc-nano](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Cloudraker | 77.23% |
| 11 | [Claude Opus 5.5](/models/claude-opus-5-5) | Proprietary VLM · Default · 5.79¢ per page | Anthropic | 77.03% |
| 12 | [aeddix-alpine-ocr (KDL pipeline)](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · KDL pipeline · no listed cost | Aeddix Labs | 76.77% |
| 13 | [florin-parser-nano](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Florin | 76.69% |
| 14 | [Nutrient DWS (Agentic)](https://www.parsebench.ai/#leaderboard) | Commercial IDP · Agentic · 1.80¢ per page | Nutrient | 76.43% |
| 15 | [Extend (2.0)](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · 2.0 · 2.50¢ per page | Extend | 75.33% |
| 16 | [Nemotron-3.5-Parse](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | NVIDIA | 75.20% |
| 17 | [LightOnOCR-3-4B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | LightOn | 75.10% |
| 18 | [Gemini 3 Flash](/models/gemini-3-flash) | Proprietary VLM · Thinking High · 2.41¢ per page | Google | 75.05% |
| 19 | [Databricks AI Parse](https://www.parsebench.ai/#leaderboard) | Commercial IDP · Default · 0.35¢ per page | Databricks | 74.71% |
| 20 | [LightOnOCR-3-0.8B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | LightOn | 74.60% |
| 21 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | Proprietary VLM · Reasoning High · 8.08¢ per page | OpenAI | 74.46% |
| 22 | [Infinity-Parser2-Pro](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | infly | 74.28% |
| 23 | [Gemini 3 Flash](/models/gemini-3-flash) | Proprietary VLM · Thinking Medium · 1.98¢ per page | Google | 74.22% |
| 24 | [Claude Opus 5.5](/models/claude-opus-5-5) | Proprietary VLM · Effort Low · 5.46¢ per page | Anthropic | 73.57% |
| 25 | [Gemini 3.5 Flash](/models/gemini-3-5-flash) | Proprietary VLM · Thinking High · 4.95¢ per page | Google | 73.36% |
| 26 | [Extend Light (1.0)](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · 1.0 · 0.63¢ per page | Extend | 73.26% |
| 27 | [Reducto (Agentic)](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Agentic · 4.76¢ per page | Reducto | 72.97% |
| 28 | [MinerU2.5-Pro-2605-1.2B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | OpenDataLab | 72.78% |
| 29 | [KDL-Frontier-Parser-nano](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | KDL AI | 72.38% |
| 30 | [Infinity-Parser2-Flash](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | infly | 72.18% |
| 31 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | Proprietary VLM · Thinking High · 3.41¢ per page | Google | 71.50% |
| 32 | [LightOnOCR-3-1B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | LightOn | 71.40% |
| 33 | [Gemini 3 Flash](/models/gemini-3-flash) | Proprietary VLM · Thinking Minimal · 0.65¢ per page | Google | 71.04% |
| 34 | [GLM-5.3-Flash](/models/glm-5-3-flash) | Proprietary VLM · Default · 0.44¢ per page | Z.AI | 70.75% |
| 35 | [Gemini 3.7 Flash](/models/gemini-3-7-flash) | Proprietary VLM · Thinking High · 1.91¢ per page | Google | 70.66% |
| 36 | [Chandra-ocr-2](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Datalab | 70.10% |
| 37 | [Gemini 3.7 Flash](/models/gemini-3-7-flash) | Proprietary VLM · Thinking Medium · 1.06¢ per page | Google | 70.09% |
| 38 | [Datalab Accurate](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Default · 1.00¢ per page | Datalab | 69.95% |
| 39 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | Proprietary VLM · Thinking Medium · 1.24¢ per page | Google | 69.90% |
| 40 | [GPT-6 Astra](/models/gpt-6-astra) | Proprietary VLM · Reasoning Low · 9.87¢ per page | OpenAI | 69.88% |
| 41 | [Claude Fable 5](/models/claude-fable) | Proprietary VLM · Default · 15.60¢ per page | Anthropic | 69.79% |
| 42 | [Qwen3.8-27B](/models/qwen3-8-27b) | Open-weight VLM · Thinking · no listed cost | Alibaba | 69.72% |
| 43 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | Proprietary VLM · Reasoning Low · 1.95¢ per page | OpenAI | 69.70% |
| 44 | [Datalab Balanced](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Default · 0.40¢ per page | Datalab | 69.51% |
| 45 | [Gemini 3.6 Flash](/models/gemini-3-6-flash) | Proprietary VLM · Thinking High · 4.25¢ per page | Google | 69.38% |
| 46 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | Proprietary VLM · Thinking Low · 0.71¢ per page | Google | 69.32% |
| 47 | [Claude Sonnet 5.5](/models/claude-sonnet-5-5) | Proprietary VLM · Default · 2.74¢ per page | Anthropic | 69.26% |
| 48 | [Gemini 3.5 Flash](/models/gemini-3-5-flash) | Proprietary VLM · Thinking Medium · 4.28¢ per page | Google | 69.20% |
| 49 | [Gemini 3.1 Pro](/models/gemini-3-1-pro) | Proprietary VLM · Default · 8.49¢ per page | Google | 69.14% |
| 50 | [GPT-6 Sol](/models/gpt-6-sol) | Proprietary VLM · Reasoning High · 3.47¢ per page | OpenAI | 68.89% |
| 51 | [Gemini 3.7 Flash](/models/gemini-3-7-flash) | Proprietary VLM · Thinking Low · 0.70¢ per page | Google | 68.71% |
| 52 | [Mistral OCR 4 (Annotation)](https://www.parsebench.ai/#leaderboard) | Commercial IDP · Annotation · 0.50¢ per page | Mistral | 68.23% |
| 53 | [Reducto](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Default · 2.38¢ per page | Reducto | 67.83% |
| 54 | [Datalab Fast](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Default · 0.40¢ per page | Datalab | 67.79% |
| 55 | [PaddleOCR-VL-1.6](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | PaddlePaddle | 67.43% |
| 56 | [GPT-6 Sol](/models/gpt-6-sol) | Proprietary VLM · Reasoning Medium · 2.40¢ per page | OpenAI | 67.38% |
| 57 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | Proprietary VLM · Reasoning Medium · 2.71¢ per page | OpenAI | 67.11% |
| 58 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | Proprietary VLM · Reasoning XHigh · 1.01¢ per page | OpenAI | 66.89% |
| 59 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | Proprietary VLM · Reasoning Max · 1.70¢ per page | OpenAI | 66.85% |
| 60 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | Proprietary VLM · Reasoning Low · 2.59¢ per page | OpenAI | 66.74% |
| 61 | [GPT-5.5](/models/gpt-5-5) | Proprietary VLM · Reasoning Medium · 13.09¢ per page | OpenAI | 66.63% |
| 62 | [Gemini 3.6 Flash](/models/gemini-3-6-flash) | Proprietary VLM · Thinking Medium · 3.76¢ per page | Google | 66.06% |
| 63 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | Proprietary VLM · Reasoning High · 0.59¢ per page | OpenAI | 66.02% |
| 64 | [PaddleOCR-VL-1.5](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | PaddlePaddle | 65.95% |
| 65 | [GPT-6 Sol](/models/gpt-6-sol) | Proprietary VLM · Reasoning None · 1.86¢ per page | OpenAI | 65.32% |
| 66 | [Surya OCR 2](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Datalab | 64.83% |
| 67 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | Open-weight VLM · Thinking High · no listed cost | DeepSeek | 64.56% |
| 68 | [GPT-6 Luna](/models/gpt-6-luna) | Proprietary VLM · Reasoning Max · 0.72¢ per page | OpenAI | 64.26% |
| 69 | [Gemini 3.6 Flash](/models/gemini-3-6-flash) | Proprietary VLM · Thinking Minimal · 1.48¢ per page | Google | 64.26% |
| 70 | [Claude Opus 4.7](/models/claude-opus-4-7) | Proprietary VLM · Default · 7.14¢ per page | Anthropic | 63.34% |
| 71 | [GPT-6 Luna](/models/gpt-6-luna) | Proprietary VLM · Reasoning XHigh · 0.29¢ per page | OpenAI | 63.15% |
| 72 | [GPT-6 Luna](/models/gpt-6-luna) | Proprietary VLM · Reasoning High · 0.21¢ per page | OpenAI | 62.95% |
| 73 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | Proprietary VLM · Reasoning None · 2.36¢ per page | OpenAI | 62.83% |
| 74 | [GPT-5.5](/models/gpt-5-5) | Proprietary VLM · Reasoning None · 5.93¢ per page | OpenAI | 62.82% |
| 75 | [HunyuanOCR 1.5](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Tencent | 62.59% |
| 76 | [Gemma 4 31B](/models/gemma-4-31b) | Open-weight VLM · Default · no listed cost | Google | 62.40% |
| 77 | [Claude Opus 4.8](/models/claude-opus-4-8) | Proprietary VLM · Default · 7.30¢ per page | Anthropic | 62.27% |
| 78 | [GPT-5.4](/models/gpt-5-4) | Proprietary VLM · Reasoning None · 2.90¢ per page | OpenAI | 62.23% |
| 79 | [Gemini 3.5 Flash](/models/gemini-3-5-flash) | Proprietary VLM · Thinking Minimal · 1.82¢ per page | Google | 61.99% |
| 80 | [Qwen3-VL-8B-Instruct](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Alibaba | 61.97% |
| 81 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | Proprietary VLM · Reasoning Medium · 0.29¢ per page | OpenAI | 61.87% |
| 82 | [Reducto (r-1)](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · r-1 · 1.00¢ per page | Reducto | 61.68% |
| 83 | [Claude Sonnet 5](/models/claude-sonnet-5) | Proprietary VLM · Default · 3.18¢ per page | Anthropic | 61.16% |
| 84 | [GPT-6 Luna](/models/gpt-6-luna) | Proprietary VLM · Reasoning Medium · 0.14¢ per page | OpenAI | 61.10% |
| 85 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | Proprietary VLM · Reasoning None · 4.06¢ per page | OpenAI | 60.86% |
| 86 | [DocAI](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Default · 1.00¢ per page | — | 60.74% |
| 87 | [Mistral OCR 4](https://www.parsebench.ai/#leaderboard) | Commercial IDP · Default · 0.40¢ per page | Mistral | 60.68% |
| 88 | [Qwen3.8-27B](/models/qwen3-8-27b) | Open-weight VLM · Default · no listed cost | Alibaba | 60.30% |
| 89 | [Gemini 3.1 Pro](/models/gemini-3-1-pro) | Proprietary VLM · Thinking Low · 2.50¢ per page | Google | 59.87% |
| 90 | [Gemini 3.5 Flash-Lite](/models/gemini-3-5-flash-lite) | Proprietary VLM · Thinking High · 1.18¢ per page | Google | 59.66% |
| 91 | [Azure Document Intelligence (Layout)](https://www.parsebench.ai/#leaderboard) | Commercial IDP · Layout · 1.00¢ per page | Microsoft | 59.64% |
| 92 | [Gemma 4 26B A4B](/models/gemma-4-26b-a4b) | Open-weight VLM · Default · no listed cost | Google | 58.50% |
| 93 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | Proprietary VLM · Reasoning Low · 0.23¢ per page | OpenAI | 58.48% |
| 94 | [Gemini 3.1 Flash-Lite](/models/gemini-3-1-flash-lite) | Proprietary VLM · Default · 0.29¢ per page | Google | 58.32% |
| 95 | [GPT-6 Luna](/models/gpt-6-luna) | Proprietary VLM · Reasoning Low · 0.11¢ per page | OpenAI | 58.03% |
| 96 | [TeleOCR](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | StarDoc AI | 57.53% |
| 97 | [Gemini 3.1 Flash-Lite](/models/gemini-3-1-flash-lite) | Proprietary VLM · Thinking Medium · 0.32¢ per page | Google | 56.83% |
| 98 | [Gemini 3.1 Flash-Lite](/models/gemini-3-1-flash-lite) | Proprietary VLM · Thinking High · 1.04¢ per page | Google | 56.28% |
| 99 | [Claude Haiku 5.5](/models/claude-haiku-5-5) | Proprietary VLM · Default · 0.14¢ per page | Anthropic | 56.10% |
| 100 | [Dots.mocr](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | rednote | 55.79% |
| 101 | [Extend](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Default · 2.50¢ per page | Extend | 55.75% |
| 102 | [Gemini 3.5 Flash-Lite](/models/gemini-3-5-flash-lite) | Proprietary VLM · Default · 0.56¢ per page | Google | 55.50% |
| 103 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | Open-weight VLM · No Thinking · no listed cost | DeepSeek | 55.20% |
| 104 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | Proprietary VLM · Reasoning None · 0.22¢ per page | OpenAI | 54.47% |
| 105 | [Claude Opus 4.6](/models/claude-opus-4-6) | Proprietary VLM · Default · 5.78¢ per page | Anthropic | 54.07% |
| 106 | [PyMuPDF4LLM](https://www.parsebench.ai/#leaderboard) | Open-source local parser · Default · no listed cost | Artifex | 53.49% |
| 107 | [HPD-Parsing](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | PaddlePaddle | 53.39% |
| 108 | [Claude Haiku 4.5 Thinking](/models/claude-haiku-4-5-thinking) | Proprietary VLM · Thinking · 3.69¢ per page | Anthropic | 53.12% |
| 109 | [Gemini 3.5 Flash-Lite](/models/gemini-3-5-flash-lite) | Proprietary VLM · Thinking Medium · 0.47¢ per page | Google | 52.22% |
| 110 | [Falcon-OCR](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | TII | 52.07% |
| 111 | [GPT-5 mini](/models/gpt-5-mini) | Proprietary VLM · Reasoning Medium · 3.05¢ per page | OpenAI | 51.52% |
| 112 | [GPT-6 Luna](/models/gpt-6-luna) | Proprietary VLM · Reasoning None · 0.09¢ per page | OpenAI | 51.40% |
| 113 | [Docling-models](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Docling | 50.65% |
| 114 | [Google Cloud Document AI](https://www.parsebench.ai/#leaderboard) | Commercial IDP · Default · 1.00¢ per page | Google | 50.39% |
| 115 | [Nemotron 3 Nano Omni 30B A3B](/models/nemotron-3-nano-omni-30b-a3b) | Open-weight VLM · Thinking · no listed cost | NVIDIA | 48.51% |
| 116 | [LightOnOCR-2-1B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | LightOn | 48.00% |
| 117 | [AWS Textract](https://www.parsebench.ai/#leaderboard) | Commercial IDP · Default · 1.50¢ per page | Amazon | 47.88% |
| 118 | [GPT-5 mini](/models/gpt-5-mini) | Proprietary VLM · Reasoning Minimal · 0.88¢ per page | OpenAI | 46.83% |
| 119 | [Qianfan-OCR](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Baidu | 46.20% |
| 120 | [Unlimited-OCR](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Baidu | 46.17% |
| 121 | [Jina-OCR-v1](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Jina AI | 45.93% |
| 122 | [MinerU2.5-2509-1.2B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | OpenDataLab | 45.90% |
| 123 | [LandingAI](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Default · 3.00¢ per page | LandingAI | 45.23% |
| 124 | [Claude Haiku 4.5](/models/claude-haiku-4-5) | Proprietary VLM · Disable Thinking · 1.64¢ per page | Anthropic | 45.17% |
| 125 | [Qwen3.6-35B-A3B](/models/qwen3-6-35b-a3b) | Open-weight VLM · Default · no listed cost | Alibaba | 44.10% |
| 126 | [GPT-5.4 nano](/models/gpt-5-4-nano) | Proprietary VLM · Default · 0.25¢ per page | OpenAI | 43.35% |
| 127 | [MinerU-Diffusion-V1-0320-2.5B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | OpenDataLab | 42.63% |
| 128 | [DeepSeek-OCR-2](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | DeepSeek | 41.20% |
| 129 | [PaddleOCR-VL](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | PaddlePaddle | 40.90% |
| 130 | [OvisOCR2](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | — | 40.84% |
| 131 | [Gemma 4 E4B](/models/gemma-4-e4b) | Open-weight VLM · Default · no listed cost | Google | 40.50% |
| 132 | [Amazon Nova 2 Lite](/models/nova-2-lite) | Proprietary VLM · Default · 0.70¢ per page | Amazon | 40.43% |
| 133 | [Warp Ingest](https://www.parsebench.ai/#leaderboard) | Open-source local parser · Default · no listed cost | Warp | 40.18% |
| 134 | [Granite Vision 4.1 4B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | IBM | 39.45% |
| 135 | [WeVisDoc-2B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Tencent | 38.93% |
| 136 | [LiteParse (no OCR)](https://www.parsebench.ai/#leaderboard) | Open-source local parser · no OCR · no listed cost | LlamaIndex | 36.90% |
| 137 | [WeVisDoc-4B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Tencent | 36.15% |
| 138 | [Qwen3.5-4B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Alibaba | 35.40% |
| 139 | [Qwen3.5-9B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Alibaba | 31.90% |
| 140 | [Firecrawl](https://www.parsebench.ai/#leaderboard) | Commercial parsing API · Default · 0.90¢ per page | Firecrawl | 31.08% |
| 141 | [GLM-OCR](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Z.AI | 29.60% |
| 142 | [OpenDataLoader](https://www.parsebench.ai/#leaderboard) | Open-source local parser · Default · no listed cost | — | 29.40% |
| 143 | [Qwen3.5-0.8B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Alibaba | 28.40% |
| 144 | [Qwen3.5-2B](https://www.parsebench.ai/#leaderboard) | Open-weight VLM · Default · no listed cost | Alibaba | 27.30% |
| 145 | [pdf-inspector](https://www.parsebench.ai/#leaderboard) | Open-source local parser · Default · no listed cost | — | 26.59% |
| 146 | [MarkItDown](https://www.parsebench.ai/#leaderboard) | Open-source local parser · Default · no listed cost | Microsoft | 18.63% |
| 147 | [PyMuPDF (HTML)](https://www.parsebench.ai/#leaderboard) | Open-source local parser · HTML · no listed cost | Artifex | 16.62% |
| 148 | [PyMuPDF (Text)](https://www.parsebench.ai/#leaderboard) | Open-source local parser · Text · no listed cost | Artifex | 16.02% |
| 149 | [pypdf](https://www.parsebench.ai/#leaderboard) | Open-source local parser · Default · no listed cost | — | 14.87% |

## ParseBench (Kaggle)

Evaluation results published by [Kaggle Benchmarks](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench). 

Source snapshot: 2026-10-23 source capture.

Published Kaggle aggregates only. Model configurations and evaluation dates stay attached to each result. These results do not enter overall or category rankings. Task data, prompts, answers, and per-case outputs are not included.

- [Kaggle leaderboard and methodology](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench)
- [Published leaderboard data](https://www.kaggle.com/api/v1/benchmarks/llamaindex-org/parsebench/leaderboard)
- [ParseBench project](https://www.parsebench.ai/)
- [Owner evaluation code and separate parser leaderboard](https://github.com/run-llama/ParseBench)
- [ParseBench paper](https://arxiv.org/abs/2604.08538)
- [Apache 2.0 license shown by Kaggle](https://www.apache.org/licenses/LICENSE-2.0)

| Source model | Configuration | ParseBench score |
|---|---|---|
| [GPT-6.1 Sol](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-10-05 | 80.1% |
| [GPT-6 Sol](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-10-01 | 80.0% |
| [Gemini 3 Flash Preview](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-16 | 79.3% |
| [GPT-6 Astra](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-09-10 | 79.1% |
| [Gemini 4 Argon](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-10-09 | 78.4% |
| [GPT-5.6 Terra](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-08-03 | 77.5% |
| [GPT-5.6 Sol](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-08-03 | 77.5% |
| [GPT-5.5](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-05-23 | 75.5% |
| [GPT-5.4](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-17 | 72.9% |
| [Gemini 3.7 Flash](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-08-26 | 69.5% |
| [Grok 4.7](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-10-02 | 69.3% |
| [Claude Opus 5](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-08-03 | 69.0% |
| [Gemini 3.8 Flash](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-09-10 | 67.8% |
| [Gemma 4 31B](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-16 | 66.4% |
| [GPT-5.6 Luna](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-08-03 | 66.0% |
| [Gemini 3.1 Pro Preview](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-17 | 64.6% |
| [Claude Opus 4.7](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-05-23 | 64.0% |
| [Claude Opus 4.8](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-05-29 | 64.0% |
| [Deepseek V4.1 Flash](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-10-20 | 63.8% |
| [Gemini 3.1 Flash-Lite Preview](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-16 | 62.0% |
| [Gemma 4 26B A4B](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-16 | 61.1% |
| [Grok 4.6](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-09-16 | 60.7% |
| [Claude Haiku 5.5](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-10-09 | 58.5% |
| [Claude Sonnet 4.6](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-16 | 56.5% |
| [Gemini 3.5 Flash-Lite](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-08-03 | 55.5% |
| [Claude Opus 4.6](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-17 | 53.8% |
| [GPT-5.4 nano](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-16 | 49.7% |
| [Gemini 2.5 Flash](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-16 | 49.2% |
| [GPT-5.4 mini](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-16 | 48.3% |
| [Gemini 3.5 Flash](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-05-23 | 34.0% |
| [Gemini 3.6 Flash](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-08-03 | 23.7% |
| [Grok 4.5](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-08-03 | 21.8% |
| [Claude Sonnet 5](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-08-03 | 3.6% |
| [DeepSeek V3.2](https://www.kaggle.com/benchmarks/llamaindex-org/parsebench) | Task version 4; Evaluated 2026-04-16 | 3.3% |

## FAQ

### What does ParseBench measure?

ParseBench measures whether a parser's output keeps what an agent needs from a page: table structure, chart values, the full text without omissions or inventions, meaningful formatting and the position of each element. Every page carries deterministic test rules, and the overall score is the mean of the five dimension scores.

### How does this table differ from the Kaggle rerun below it?

This table mirrors LlamaIndex's own leaderboard: 148 configurations including commercial parsing APIs, open-weight OCR models and local parsers. Kaggle Benchmarks reruns the same task on 33 models through its own harness and publishes different numbers for the same models, so its board is shown as a separate table on this page and never merged.

### Why is ParseBench display only?

Each row is one configuration under LlamaIndex's harness, reasoning settings change both score and cost, and the publisher sells the leading system. The board is public and reproducible, which is why we show every row, but the results compare those exact runs and do not enter BenchLM's model rankings.